<ul><li>An independent team has developed a Pokémon model with just 10 million parameters, significantly smaller than frontier AI models.</li><li>The small model outperforms larger models due to the 'depth curse', a counterintuitive issue in AI.</li><li>The model was trained using Reinforcement Learning algorithm and achieved the goal efficiently.</li><li>The article discusses imitation learning and exploration learning as two training paradigms in AI.</li></ul>

60,500 Times Smaller, but Better; AI’s Depth Curse.

Discover more